Yup, I understand the general concept of preference falsification. My question is about the specific application. I think it would be helpful if you had a concrete example of where this would be relevant for e.g. malaria bednets or factory farming?
Larks
Karma: 16,115
Thanks for sharing this. Perhaps you could explain the relevance to effective altruism a bit more explicitly?
That mindspace is large and AIs are really weird.
That it is possible to make smarter than human AIs and that this is The Issue.
That sufficiently intelligent AIs will not ‘automatically’ be moral (e.g. the behaviour of un-RLHF’d models).
That people will choose to let the AI out of the box.
since criticisms of EA also come with the ‘Community’ tag attached
This seems not straightforwardly true to me—or at least it shouldn’t be? Criticisms of the EA community should be community-tagged, but criticisms of EA ideas should not be.
This is a guy who got back surgery that was covered by his health insurance and then murdered the CEO of a different health insurance company. While EAs are always keen to self-flagellate over any possible bad thing that might have some tangential connection, I really think this one can be categorized under ‘crazy’ and ‘psychedelics’. To the extent he was motivated by ideology it doesn’t seem to be EA—the slogan he carved onto the bullet casings was a general anti-capitalist anti-insurance one.
Surely smokers are on average richer than malaria victims? No-one in the west is getting Malaria, but many do smoke, and smokers need at least enough money to buy cigarettes. And smokers chose to smoke; people aren’t choosing to get malaria.
unlike many problems the EA community grapples with, we already know how to help them
It’s strange you make this post about comparing to malaria. Surely we actually have a better sense of how to fight malaria (e.g. bednets, medicine) than we do smoking (where even with good support quitting can be difficult)?
Even considering that you could presumably choose not to apply? (I guess you think it is bad in a systematically surprising way).
Presumably you would want to give them negative rights (contracts, own property, not being murdered etc.) only, and not positive rights (healthcare, housing, cash). Which makes sense to me as the latter are much more philosophically dubious anyway.
So technically, they received some aid—I’ll edit accordingly, thanks for the flag—but considerably less than most refugees.
There is a huge difference between ’they were at some times not approved for this specific type of aid” and “work or starve”. There is no way that the US in the 1980s would tolerate mass starvation like this—even if the federal government hadn’t stepped in, the individual states, churches, charities, families etc. would not have allowed that to occur.
To quote Billy and Packard 2020...
If you read the prior sentence in that article, you will see they are basically assuming the negative selection to be true, and don’t engage with my argument that positive selection effects also existed at all:
Accounting for migrant selection lies outside the scope of our project and the available data.
I don’t think the fact that some were eventually deported shows very much. I’m not denying that some of them were criminals—I’m just claiming that there are also significant positive selection effects. Since you’re not saying that they were all eventually deported, and I’m not saying that every single migrant was a great person, I don’t think the mere fact that some were deported is very strong evidence either way.
That’s their… headline result?
No, it is not. You discussed whether refugees were “particularly likely to commit crimes”. This is a simple statistic—you take crimes committed and divide by population. It is the statistic shown in the chart I included. As far as I am aware, basically every source agrees that this wave of refugees commit crimes at well above the rates of natives.
In contrast, my understanding is the Huang and Kvasnicka paper you quoted do a series of regressions to try to establish whether the scale of immigration changed the amount of crimes that refugees committed. This is a different question. It could (hypothetically) be the case that refugees were committing crimes at a very high rate, and then this fell in 2015 but was still higher than the native rate—if this was the case then this paper would show the opposite result to what we are discussing.
I am also very skeptical of the paper because the garden of branching paths issue seems so large—they declined to publish simple statistics and opted for much more complicated regressions instead which matched the results they clearly ideologically favoured—but this is beside the point because, even if their paper had no issues, it simply answers a different question.
It’s possible I’ve misunderstood this issue. If that’s the case I’d love to see the explanation for the difference between this paper’s complex methodology and the simple approaches which overwhelmingly suggest the opposite.
Marielitos were also ineligible for government assistance
Are you sure this is true? I think both state and federal government provided a lot of aid, as is typical for refugees. See for example President Carter’s speech:
This legislation means that $100 million will immediately be available to the communities of Florida and a few other States to help reimburse them for expenses involved in the recent influx of people from Cuba and from Haiti.
...
Congress has already appropriated $100 million in reimbursements for cash and medical assistance and social services provided to these newcomers; this amendment makes that money available.I am also skeptical of this inference you make:
Especially unusually, Mariel immigrants were even sometimes negatively selected—people Castro wanted to get rid of
Castro was a communist dictator. While some of the people leaving I would expect to be criminals, I would also expect those who were opposed to communism and wanted valued the opportunities and freedoms offered by capitalism to be highly represented. If you wanted to work hard and better yourself, why would you not want to move from Cuba to the US? I would expect them to be significantly positively skewed, especially compared to recent refugees into Europe, who have access to generous government benefits.
I am also confused by your claim here:
In Germany, refugees were not particularly likely to commit crimes against Germans
Despite looking through the paper you cite (and consulting Notebook LM), I could not actually find this comparison in it. It is clear the authors have the data to calculate it—they have crimes tagged by whether they were committed by a refugee or not—but curiously they seem to have forgotten to actually calculate the ratio, instead opting only to show more complicated regression results whose conclusions, I imagine, were more agreeable to them.
However, the basic statistics are easily available elsewhere, despite the German state’s attempts to suppress reporting on the subject. Even the BBC publishes them:
It seems like this should be normalized by total population. If a country only had one depressed dude, and he was untreated, I would say this is a small gap, but the map as it is would suggest it was the largest possible gap. Conversely, if every single person in the country was depressed, and only 33% were treated, this map would suggest the gap was very small.
Gwern on creating your own AI race and China’s Fast Follower strategy.
I remember removing an org entirely because they complained, though in that case they claimed they didn’t have enough time to engage with me (rather than the opposite). It’s also possible there are other cases I have forgotten. To your point, I have no objections to Michael’s “make me overly concerned about being nice” argument which I do think is true.
It takes a lot longer. I reviewed 28 orgs; it would take me a long time to send 28 emails and communicate with potentially 28 people.
This is quite a scalable activity. When I used to do this, I had a spreadsheet to keep track, generated emails from a template, and had very little back and forth—orgs just saw a draft of their section, had a few days to comment, and then I might or might not take their feedback into account.
Here, it seems reasonable to assume that orgs will have made a conscious decision about what general information they want to share with would-be small/medium donors. So there isn’t much reason to expect that an inquiry (along with notice that the author is planning to publish on-Forum) would yield material additional information.[1]
This seems quite false to me. Far from “isn’t much reason”, we already know that such an inquiry would have yielded additional information, because Malo almost definitely would have corrected Michael’s material misunderstanding about MIRI’s work.
Additionally, my experience of writing similar posts is that there are often many material small facts that small orgs haven’t disclosed but would happily explain in an email. Even basic facts like “what publications have you produced this year” would be impossible to determine otherwise. Small orgs just aren’t that strategic about what they disclose!
These seem like poor things to bet on:
Trump will commit more than x crimes during his presidency.
This lacks an objective resolution criteria, and ‘number of crimes’ in the US is often a fairly random number because a single act can give rise to multiple violations. Also, committing crimes is very different from being an autocrat—you could be an autocrat and obey the law, and you can be a democrat and break the law.
Trumps secretaries will commit more than x crimes during his presidency
Similar issues.
Trump will attempt to run for a third term
Not as bad, but seems insufficient. Michael Bloomberg ran for a third term as NYC mayor, even though this required changing the rules just for him, but he was not an autocrat.
The winner of the republican primary in the next two presidential elections will be a MAGA
This is subjective, and also insufficient, as whatever ‘MAGA’ is, it is not the same as an autocrat.
In the next x years, a future president or (sufficiently) high up politician will not be convicted of any crimes conditional on their party controlling the justice department
This also seems insufficient to demonstrate autocracy—for example to my knowledge Obama was never convicted of any crimes when his party controlled the Justice Department, but he was not an autocrat.
I think the best thing to bet on is the probability of winning the next election. Unfortunately this doesn’t work nearly as well as it would have a few weeks ago, but I think think it is the best approach.
Conversations people have with un-RLHF’d models.